lgb.wrap.loss: LightGBM loss function wrapper

In Laurae2/LauraeDS: Laurae's Data Science Package

Description

The wrapper works only if both the wrapper and the original loss metric are existing. Requires Matrix and lightgbm packages.

Usage

lgb.wrap.loss(f)

Arguments

f	Type: function. The function to wrap from LightGBM. Requires the following order of arguments for the function to work: preds, labels, and returns a vector of the same length of both the inputs.

Value

The wrapping function.

Examples

# Note: this example unexpectedly fails when using pkgdown.

library(lightgbm)
library(Matrix)
data(agaricus.train, package = "lightgbm")
train <- agaricus.train
dtrain <- lgb.Dataset(train$data, label = train$label)

cross_entropy <- function(preds, labels) {
  preds <- 1 / (1 + exp(-preds))
  grad <- preds - labels
  hess <- preds * (1 - preds)
  return(list(grad = grad, hess = hess))
}
cross_entropy_wrap <- lgb.wrap.loss(f = cross_entropy)
params <- list(learning_rate = 1, min_data = 1, nthread = 1)

set.seed(1)
model <- lgb.cv(params,
                dtrain,
                2,
                nfold = 5,
                obj = cross_entropy_wrap,
                metric = "auc")

Laurae2/LauraeDS documentation built on May 29, 2019, 2:25 p.m.